← All episodes

How agents will change banking forever | E2260

| 7 products mentioned
Watch on YouTube ai agents banking automation self-improving models job displacement agentic workflows social contract phone automation

This episode explores how AI agents are fundamentally transforming banking, business automation, and employment while highlighting a stark divide in how different countries perceive AI's future. The hosts discuss Andre Karpathy's self-improving LLM framework, demonstrate real-world agent applications in financial services and mobile automation, and analyze why Americans distrust AI far more than citizens in China—revealing deep concerns about broken social contracts in corporate America that threaten political backlash against the technology industry.

Key takeaways
  • Self-improving AI models can now iteratively enhance their own code and performance through simple recursive loops, democratizing AI development beyond elite researchers and enabling non-experts to experiment with model optimization.
  • Americans' distrust of AI stems from a broken social contract where corporate profits now rise through headcount cuts and offshoring rather than increased wages and hiring, making citizens rationally skeptical of automation despite technological progress.
  • Agentic banking systems with cryptographic security and user approval workflows enable AI to manage financial tasks like bill payments and account optimization while maintaining human control through biometric-secured transaction approval.
  • Workers can outrun automation by moving up the skill stack toward jobs robots cannot yet perform—trades like carpentry, plumbing, and electrical work currently command $75-100/hour versus $20-40/hour gig economy work threatened by automation.
  • Phone automation agents using vision models can now control Android devices and orchestrate multi-step workflows like social media posting across parallel devices, opening new possibilities for competitive research, QA testing, and workflow automation.
  • The recursive self-improvement framework Eugene Stuckless demonstrated for AI-native QA testing shows how agents can optimize their own workflows across multiple layers while maintaining governance controls and blast radius isolation to prevent prompt injection vulnerabilities.
  • Super-distribution strategies—posting content multiple times across different platforms in various formats through AI assistants—dramatically increase productivity for founders and operators managing social presence at scale.

Recommendations (6)

Abby Jane recommends

"I like a certain bakery in the Austin area. It's called Abby Jane. They make incredible bread. They use the whole grains, like artisanal grains. So it's just this is literally the best bakery in Te..."

This Week in Startups · ▶ 31:15

Uber
Uber uses

"Uber has a courier service. The courier service is like 10 bucks, but you have to coordinate. So, what do we do? My Athena assistant looks at their menu at 7 a.m., finds out what's new on the menu ..."

This Week in Startups · ▶ 32:08

Auto Research recommends

"What this is it's a really stripped down LLM training loop and it runs in five-minute increments. So you bring your own AI model to be an agent essentially and then you give it a prompt and then wh..."

The Game w/ Alex Hormozi · ▶ 1:01

Claude
Claude uses

"being able to use Claude to work to make a reoccurring job"

This Week in Startups · ▶ 5:12

OpenClaw
OpenClaw uses

"being able to install OpenClaw, being able to write a skill, maybe publish it to GitHub"

This Week in Startups · ▶ 5:15

ChatGPT
ChatGPT uses

"I'm gonna take out YouTube. I'm gonna take out ChatGPT. I'm gonna learn how to be a carpenter"

This Week in Startups · ▶ 26:33

Mentioned (1)

OpenAI
OpenAI "He worked at OpenAI, I think, twice. He was like in between stints at Tesla or maybe Tesla was be..." ▶ 1:44